Introduction to Deep Learning
Discover Deep learning and find out what it is.
Deep Learning is widely used in the modern world. It is applied everywhere, from healthcare to manufacture. It helps to find a solution of some complicated issues like speech and object recognition, computer-aided translation etc.
One of the most impressive events of the last year was the victory of AlphaGo, a computer program having defeated the best Go player in the world. Besides, computers have had the upper hand in many other games: chess, checks, Reversi and Jeopardy.
Perhaps it seems that a victory in a board game has nothing in common with finding a solution of a real problem, but it is not quite true. Go was created the way AI had no chances to win the game. There was something important it lacked to get ahead: human intuition. Now finally there is a solution allowing the artificial intelligence to face many issues it was unable to deal with earlier.
It’s obvious that Deep Learning is still far from being perfect but nevertheless close to become commercially beneficial. For example, self-driving cars. Famous companies like Google, Tesla and Uber are already trying to introduce them into everyday life.
Ford estimates that by 2021 the growth of development of self-driving transport will become even more significant. The government of the USA has already developed traffic safety rules for it.
What is Deep Learning?
To answer this question, you need to understand how it interacts with Machine Learning, neural nets and AI. The scheme of the interaction is visually represented below as a set of concentric circles:
The external circle represents the artificial intelligence as a general notion (including, for example, computers). A bit further, you can see Machine Learning and in the center there are Deep Learning and artificial neural nets.
Broadly speaking, Deep Learning is just a more accurate term for artificial neural nets. Here «deep» stands for the degree of complexity (depth) of a neural net, that can often be rather superficial.
The developers of the first neural net were inspired by the structure of brain cortex. The basic level of the net, the perceptron, is actually a mathematical counterpart of a biological neurone. Mutually intersected perceptrons may appear both in the human brain and in a neural net.
The first neural net layer is called the input layer. Each node of the layer receives some input information and transmits it to the following nodes situated in other layers. The most often, there is no connection between the nodes of the same layer and the last node of the chain gives out the result of the work of the neural net.
Middle nodes are called hidden because unlike input and output nodes they are not connected in any way with the outer world. They are activated only if all the superior layers are active.
Deep Learning is a learning method involving a number of layers and used by a neural net where it is necessary to perform complicated pattern-based tasks. In particular, it is applied for speech recognition. In the eighties, the biggest part of neural nets had only one layer because they were expensive in use and abilities of the data a neural net worked with were limited.
If Machine Learning is often considered as a branch or a variation of AI, Deep Learning is a special type of the above variation.
Machine Learning uses AI, which takes time to give a response. Instead, the code will be activated using test data and the code stroke will be adjusted accordingly, depending on whether the data results are correct or not. There is a number of techniques and special software applied when performing this process, as well as computer science literature describing static methods and linear algebra used to obtain a positive result.
In the next part of the article you will find out about the importance of deep learning, its methods and deep learning related microservices. This article will also provide you with lots of resources helping you to get familiar with the notion.